Enhanced Alzheimer's disease and Frontotemporal Dementia EEG Detection: Combining lightGBM Gradient Boosting with Complexity Features

Alzheimer's disease and Frontotemporal dementia are the two most reported dementia cases. They both are neurodegenerative disorders without cure while existing treatments only halt their progress. Thus, early detection is of crucial importance. In this work, we utilize electroencephalographic signals of AD and FTD patients and propose a classification pipeline to distinguish them from healthy signals. This pipeline consists of Independent Component Analysis as a preprocessing stage, the extraction of time, frequency and complexity features, feature elimination through importance ranking and finally classification through utilizing Gradient Boosting Decision Trees. The proposed methodology achieved 92.27% F1 score in the Dementia versus Control problem, 83.06% in the AD versus Control and 80.67% in the FTD versus Control.

I. Introduction

Dementia is a wide group of neurodegenerative brain disorders that leads to cognitive impairment and interferes with social and occupational functioning [1]. The occurrence of dementia increases in frequency with age and its treatment is based on therapies that slow the progress of the illness or manage symptoms. Alzheimer's disease (AD) is the most common type of dementia and it involves severe memory loss, cognitive impairment and changes in behavioral changes [2]. Another common type of dementia is the Frontotemporal (FTD), which causes focal degeneration in the frontal lobe, anterior temporal lobe and islet.